Code repo for SRDAN, Scale-aware and Range-aware Domain Adaptation Network for Cross-dataset 3D Object Detection (CVPR2021). The code is implemented based on OpenPCDet. Please see README_SRDAN for more details.
Datasets used:
- Nuscenes (Real dataset, cross-scene)
Example: After Download whole dataset, Create Nuscene db and Run the training. Please refer to OpenPCDet and README_SRDAN for additional dataset processing issues.
- cp modify_nusc_lib/splits.py ~/PCDet/lib/python3.6/site-packages/nuscenes/utils/splits.py
- cp modify_nusc_lib/loaders.py ~/PCDet/lib/python3.6/site-packages/nuscenes/eval/detection/loaders.py
- python pcdet/datasets/nuscenes/nuscenes_dataset.py create_nuscenes_infos boston
- python pcdet/datasets/nuscenes/nuscenes_dataset.py create_nuscenes_dbinfos boston
- Astar3D(Real dataset, day-to-night)
- Kitti (Real dataset, synthetic-to-real)
- PreSil (Synthetic dataset, synthetic-to-real)
Please consider cite:
@InProceedings{Zhang_2021_CVPR,
author = {Zhang, Weichen and Li, Wen and Xu, Dong},
title = {SRDAN: Scale-Aware and Range-Aware Domain Adaptation Network for Cross-Dataset 3D Object Detection},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2021},
pages = {6769-6779}
}